Vol. 23 no. 18 2007, pages 2385–2390BIOINFORMATICS ORIGINAL PAPER doi:10.1093/bioinformatics/btm360

Gene expression

AffyProbeMiner: a web resource for computing or retrieving

accurately redefined Affymetrix probe setsHongfang Liu1,2,*,y, Barry R. Zeeberg1,y, Gang Qu3, A. Gunes Koru4,Alessandro Ferrucci1,5, Ari Kahn1,6, Michael C. Ryan6, Antej Nuhanovic7,8,Peter J. Munson7, William C. Reinhold1, David W. Kane8 and John N. Weinstein11Genomics and Bioinformatics Group, Laboratory of Molecular Pharmacology, Center for Cancer Research, NationalCancer Institute, National Institutes of Health, Bethesda, MD 20892, 2Department of Biostatistics, Bioinformatics, andBiomathematics, Georgetown University Medical Center, 4000 Reservoir Road, NW, Washington, DC 20007,3Electrical & Computer Engineering and Institute for Advanced Studies, University of Maryland, College Park, CollegePark, MD 20742, 4Department of Information Systems, 5Department of Computer Science and Electrical Engineering,University of Maryland, Baltimore County, 1000 Hilltop Circle, MD 21050, 6Department of Bioinformatics, GeorgeMason University, Fairfax, Virginia, 20110, 7Mathematical and Statistical Computing Laboratory, Division ofComputational Bioscience, Center for Information Technology, National Institutes of Health, Bethesda, MD 20892 and8SRA International, 4300 Fair Lakes Court, Fairfax, VA 22033, USA

Received on February 1, 2007; revised on June 20, 2007; accepted on July 9, 2007

Advance Access publication July 27, 2007

Associate Editor: Joaquin Dopazo

ABSTRACT

Motivation: Affymetrix microarrays are widely used to measure

global expression of mRNA transcripts. That technology is based

on the concept of a probe set. Individual probes within a probe

set were originally designated by Affymetrix to hybridize with the

same unique mRNA transcript. Because of increasing accuracy in

knowledge of genomic sequences, however, a substantial number of

the manufacturer’s original probe groupings and mappings are now

known to be inaccurate and must be corrected. Otherwise, analysis

and interpretation of an Affymetrix microarray experiment will be

in error.

Results: AffyProbeMiner is a computationally efficient platform-

independent tool that uses all RefSeq mature RNA protein coding

transcripts and validated complete coding sequences in GenBank

to (1) regroup the individual probes into consistent probe sets and

(2) remap the probe sets to the correct sets of mRNA transcripts.

The individual probes are grouped into probe sets that are

‘transcript-consistent’ in that they hybridize to the same mRNA

transcript (or transcripts) and, therefore, measure the same entity

(or entities). About 65.6% of the probe sets on the HG-U133A chip

were affected by the remapping. Pre-computed regrouped and

remapped probe sets for many Affymetrix microarrays are made

freely available at the AffyProbeMiner web site. Alternatively, we

provide a web service that enables the user to perform the

remapping for any type of short-oligo commercial or custom array

that has an Affymetrix-format Chip Definition File (CDF). Important

features that differentiate AffyProbeMiner from other approaches are

flexibility in the handling of splice variants, computational efficiency,

extensibility, customizability and user-friendliness of the interface.

Availability: The web interface and software (GPL open source

license), are publicly-accessible at http://discover.nci.nih.gov/

affyprobeminer.

Contact: hl224@georgetown.edu or barry@discover.nci.nih.gov

1 INTRODUCTION

Microarrays are widely used for large-scale, quantitative

molecular profiling of biological specimens (Stoughton, 2005).

However, because of poor reproducibility across different

platforms or different generations of the same platform(Carter et al., 2005; Kong et al., 2005; MAQC Consortium,

2006), the validity of microarray results remains a subject of

concern to the scientific community. Because hybridization to amicroarray is sequence-dependent, it is affected by mis-

assignment of probes, ambiguous assignment of probes, and

alternative splicing of target transcripts. With respect to the

latter, the percentage of genes exhibiting alternative splicing ishigh—variously estimated as 30 and 99% (Boue et al., 2003;

Lee and Roy, 2004). Accurate quantitation requires knowledge

of both the identity of the genes and the splice forms that are

expressed. Ideally, probes in a probe set should all target thesame set of transcripts. With improved genomic knowledge,

we have become increasingly aware of the deviation from that

ideal. Table 1 shows examples of the two major classes of

mapping problems that necessitate redefinition of probe sets onthe Affymetrix HG-U133A chip:

(1) A probe set contains some probes that match multipletranscripts—Probes within a probe set do not all target

the same set of transcripts. The expression levels

measured by those probes will introduce inconsistencies

yThe authors wish it to be known that, in their opinion, the first twoauthors should be regarded as joint First Authors.

� 2007 The Author(s)This is an Open Access article distributed under the terms of the Creative Commons Attribution Non-Commercial License (http://creativecommons.org/licenses/by-nc/2.0/uk/) which permits unrestricted non-commercial use, distribution, and reproduction in any medium, provided the original work is properly cited.

by guest on May 29, 2013

http://bioinformatics.oxfordjournals.org/

Dow

nloaded from

in quantitation algorithms. Table 1 illustrates the type of

complication that can arise:

� Affymetrix had originally represented the human genes

CLEC2D by one probe set, 220132_s_at, and NPM1

by two probe sets, 221691_x_at and 200063_s_at.

� Currently, three RefSeqs represent CLEC2D, and three

RefSeqs represent NPM1.

� The entries in Table 1 for each probe set (column)

identify the probes that match the RefSeqs (rows). For

example, all 11 probes in probe set 220132_s_at match

NM_013269.

� The level of hybridization to probe set 200063_s_at

provides a consistent estimate of the composite

expression for RefSeqs NM_002520 and NM_199185

of NPM1. None of the probes in the set hybridize with

RefSeq NM_001037738. However, expression of

RefSeq NM_001037738 is reflected in the hybridization

of probe set 221923_s_at.

� In contrast, if we use probe set 221691_x_at to measure

the expression of transcripts of NPM1, the level of

hybridization to the probe set could reflect cross-

hybridization with RefSeqs of CLEC2D.

(2) Some probes in a probe set do not match the target

transcript(s)—Several probes within a probe set may not

match any of the transcripts for the gene that Affymetrix

had originally designated for the probe set. The expres-

sion levels measured by those probes do not reflect the

composite expression of the transcripts of the intended

gene and therefore introduce an inconsistency in quanti-

tation. For example, again in relation to Table 1:

� Probes 7 and 8 of 221691_x_at do not target

NM_199185, which represents NPM1, but they do

target all three transcripts of CLEC2D.

� Therefore, the expression levels measured by

221691_x_at do not consistently reflect the composite

expression of the RefSeqs of the intended gene.

After the probe sequence information was made public by

Affymetrix, several publications addressed those mapping

problems by redefining probe sets (Carter et al., 2005;

Dai et al., 2005; Gautier et al., 2004; Harbig et al., 2005;

Kong et al., 2005). For example, Harbig (Harbig et al., 2005)used BLAST to match probes with documented and postulatedhuman transcripts, and redefined about 37% of the probes on

the HG-U133 plus 2.0 array. They found that the originalAffymetrix annotation was compromised because of thepotential for cross-hybridization with splice variants or

transcripts of other genes that contained matching sequences.More than 5000 probe sets were shown to hybridize withmultiple transcripts. They proposed a sequence-based identifi-

cation method in which probe sets were remapped to the mostclosely related RefSeqs. An R package, altcdfenvs, wasdeveloped by Gautier (Gautier et al., 2004) to provide

alternative mapping of probes. As proof of concept, thepackage was applied to those genes that are represented inRefSeq. Probes that matched multiple RefSeqs were eliminated

from consideration. Dai (Dai et al., 2005) recently providedredefinitions under several conditions. They suggested that theoriginal Affymetrix probe set definitions are imperfect, thatconclusions derived from past Affymetrix GeneChips may be

somewhat inaccurate, and that researchers should re-analyzetheir data.Probe sequence information was also used to address the

problem of poor reproducibility across different platforms ordifferent generations of the same platform. For example, Carter(2005) redefined Affymetrix probe sets by sequence overlap

with cDNA microarray probes to reduce cross-platforminconsistencies in cancer-associated gene expression measure-ments. Their study implied that ‘probes targeting identical

transcript sequence regions give substantially strongerconcordance than probes that target identical contiguoustranscript molecules at different sequence regions’. It also

suggested that discrepancies among different platformsare caused by improper cross-platform probe matching.Kong (2005) used sequence information to increase the

compatibility among different generations of Affymetrixarrays. They filtered probes that were not consistent with theAffymetrix annotation.

Despite those positive developments, no computationally-efficient, broadly-capable tools for remapping have been madeavailable to date. The only publicly available tool, the

BioConductor package altcdfenvs, is written in R.Unfortunately, R suffers from poor memory management(Hornik, 2007): all objects are kept in memory until garbage

collection is automatically invoked. For example, the match-probes package (used in altcdfenvs) took several days to alignprobes with mRNA sequences (Elo et al., 2005). We tried to use

the altcdfenvs package to obtain remapped Chip DefinitionFiles (CDFs) for HG-U133A. The package could be executedsuccessfully on a Unix server with a test set of 500 human

RefSeqs. However, when we tested the complete set of humanRefSeqs, no results were obtained even after 10 days ofprocessing.

To address such problems, we have developedAffyProbeMiner, a computationally efficient and platform-independent tool. It uses mRNA RefSeqs and validated

complete coding sequences in GenBank to (1) regroup theindividual probes into consistent probe sets and (2) remap theprobe sets to the correct sets of mRNA transcripts.

AffyProbeMiner uses a local implementation of the UCSC

Table 1. Example of ambiguous and mismatched probes in the

original assignments for the Affymetrix HG-U133A chip

Gene RefSeq Probe set

220132_s_at 221691_x_at 200063_s_at 221923_s_at

CLEC2D NM_013269 1-11 1,3,4,7,8 – –

NM_001004419 1-11 1,3,4,7,8 – –

NM_001004420 1-11 1,3,4,7,8 – –

NPM1 NM_002520 – 1–9 1–11 –

NM_199185 – 1–6,9 1–11 –

NM_001037738 – 1–11 – 1–11

H.Liu et al.

2386

by guest on May 29, 2013

http://bioinformatics.oxfordjournals.org/

Dow

nloaded from

BLAT server (Kent, 2002) to circumvent the mapping bottle-

neck of altcdfenvs.

2 METHODS

We define a transcript-consistent probe set as one in which each probe

maps to the same set of transcripts and a transcript-unique probe set as a

transcript-consistent probe set in which each probe maps to a single

transcript. A transcript-consistent probe set has the advantage of

eliminating the inaccuracy that would have resulted from mixing probes

that match disjoint sets of transcripts. A transcript-unique probe set has

the additional advantage of eliminating the ambiguity inherent in

lumping together the measurements of multiple transcripts. Ideally, we

would eliminate from consideration probe sets that were not transcript-

unique. Unfortunately, that degree of stringency would give up too

much of the potential information. We are not willing, however,

to accept probe sets that are not transcript-consistent because they

produce estimates that are not merely ambiguous, but are also

incorrect.

We also define the weaker concepts of gene-consistent and gene-

unique probe sets. A gene-consistent probe set is one in which each

probe maps to transcripts of the same set of genes, and a gene-unique

probe set is one in which each probe maps to transcripts of a single

gene. That is, the individual probes in gene-consistent and gene-unique

probe sets may target different splice variants of the gene(s) in question.

We reluctantly define and support resources in AffyProbeMiner for

those users who choose to ignore the implications of splice variation.

However, we suggest that, when possible, microarray data analysis be

performed at the transcript level to avoid inconsistency in the results

and their interpretation.

After AffyProbeMiner processing, the data are returned to the

normal Bioconductor or Affymetrix stream for completion of the

analysis using MAS5, RMA, or some other algorithm (Breitling, 2006).

The following section provides a brief description of the methods as

follows:

Construction of a database containing validated complete

coding sequences from GenBank and RefSeq: We obtained a list of

the raw complete coding sequences in GenBank. Those sequences

could be detected because their GenBank records contained ‘complete

CDS’ or ‘complete sequences’. We excluded sequences whose

GenBank records contained ‘intronic transcript’ or ‘BAC clone’. All

RefSeq records having accessions starting with ‘NM_’ (i.e., mature

RNA protein coding transcripts) were included in the list.

Because of the variable reliability of records in GenBank,

we determined their validity by alignment to the current genome

builds. We considered a record to be valid if �95% of the sequence can

be aligned with the genome with �99% identity. Figure 1 shows a

typical two-dimensional distribution representing those two criteria

applied to human GenBank sequences. Based on those two criteria,

94.6% (46 546 out of 49 189) complete coding sequences in GenBank

were considered to be valid.

Sometimes the ends of a valid sequence contain vector contamination

or a poly(A) tail. We detected those during BLAT alignment to the

genome sequences and removed the unmatched ends. The cleaned,

validated records were then de-duplicated. Records were considered to

be replicates if (1) NCBI annotated them as mapping to the same gene,

(2) the coding regions have the same length and (3) one sequence can be

subsumed by the other sequence with differences in nucleotides identity

�1%. If GenBank and RefSeq records were found to be duplicates, we

retained the RefSeq record. In the case of two duplicate GenBank

records, we kept the longer one. The result was a validated non-

redundant complete coding sequence database.

Generation of redefined CDFs containing re-mapped and re-grouped

probes: We downloaded all of the probe sequence files from the

Affymetrix website. The mappings between probe sequences and

complete coding sequences were determined by BLAT. The redefinition

of CDFs was based on the mapping results. AffyProbeMiner

circumvents the bottleneck in the remapping process that occurs

when the computations are performed in the R language. The

processing flow diagram (Fig. 2) indicates how it does so (red font),

to provide remapped CDF files that are suitable for various microarray

data analyses.

AffyProbeMiner’s suite of Perl programs for generating CDFs

is downloadable at the project website, http://discover.nci.nih.gov/

AffyProbeMiner. The public availability of the programs allows a user

to obtain CDFs with her/his own parameter settings. Advanced

users can customize the parameter settings or program code.

The FAQ and User Guide at the AffyProbeMiner website

are applicable to both transcript-level and gene-level analyses.

The User Guide provides detailed instructions for using our remapped

CDF files in the Affymetrix GCOS, dCHIP, or BioConductor

environments.

In addition to examining all probe sets (‘condition A’), we

also constrained the study to those probe sets having at least five

probes (‘condition F’). We believe that, as a heuristic rule of thumb,

a minimum of five probes should be required in a probe set to produce

reliable measurements. The statistical characterization is similar

whether no threshold or the five-probe threshold is used. For the

human HG-U133A chip, the probes that match a subsequence of a

complete coding sequence comprise a significant fraction (206 624/247

966� 100%¼ 83.3% and 187 232/247 966� 100%¼ 75.5% under

conditions A and F, respectively) of the probes on the chip.

Fig. 1. Two-dimensional distribution as a function of percent query

length and percent identity. All human complete coding sequences in

GenBank were aligned to the human genome. The output data of

BLAT were parsed to determine the percent of the query length that

aligned to the genome and the percent identity of the aligned portion of

the query sequence. The rectangular box in the lower right corner

encloses the region of acceptance. 94.6% of the human complete coding

sequences met the acceptance criteria. The distribution values are given

as log 10. Original distribution values were incremented by 0.10 to

avoid attempting to compute log (0).

AffyProbeMiner: a web resource for Affymetrix

2387

by guest on May 29, 2013

http://bioinformatics.oxfordjournals.org/

Dow

nloaded from

Additionally, the majority of the probes (197 249/247

966� 100%¼ 79.5% and 196 695/247 966� 100%¼ 79.3%) are gene-

unique, whereas only a small fraction (69 344/247 966� 100%¼ 27.9%

and 66 213/247 966� 100%¼ 26.7%) are transcript-unique.

3 RESULTS AND DISCUSSION

Affymetrix microarrays are widely used to measure global

expression of mRNA transcripts. Overall analysis of an

Affymetrix microarray experiment requires a combination of

specific probe expression results from the experiment and a pre-

defined CDF that groups probes into probe sets. Popular

software systems for analysis of the expression data within the

context of a given CDF are Bioconductor R (Gentleman et al.,2004), Affymetrix GCOS, and dChip. All of the individual

probes within a given probe set were originally selected by

Affymetrix to hybridize with the same unique mRNA

transcript. However, because of increasing accuracy in our

knowledge of genomic sequences (particularly for the human

genome), a substantial number of the manufacturer’s original

probe groupings and mappings need to be corrected. Analysis

and interpretation of an Affymetrix microarray experiment will

be in error both qualitatively and quantitatively in the absence

of corrected regrouping and remapping.For the remapping process (Fig. 2), we have developed

AffyProbeMiner, a computationally efficient platform-

independent tool that uses all mRNA RefSeqs and complete

coding sequences in GenBank to (1) regroup the individual

probes into consistent probe sets and (2) remap the probe sets

to the correct sets of mRNA transcripts.There are, in total, 48 Affymetrix chips for 25 organisms.

Table 2 shows the number of raw complete coding sequences

extracted for each organism when considering RefSeq only and

when considering both RefSeq and GenBank. We constructed a

complete coding sequence database for every organism

represented by an Affymetrix chip as of January 1, 2006 and

having �5000 raw complete coding sequences (RefSeq or

GenBank). We generated redefined CDFs for each chip

associated with those organisms. We considered only perfect

matches in the pre-computed results, and remapped only probe

sets for which at least five probes were retained. All of those

processing steps are fully automated using Perl and R scripts,

and it takes approximately 20 h to build, from scratch,

the complete release of AffyProbeMiner’s remapped and

redefined CDFs for 64 microarray sets representing 9 species.

The processing step that relieves the bottleneck in R takes

approximately 3min/chip (PowerBook G4 with 1G memory).

The pre-computed, remapped probe sets are freely available to

the scientific community via download from our web site.

Alternatively, the computation can be performed as a web

service on our server either for Affymetrix arrays or for any

other short-oligo that use Affymetrix-format CDFs. The web

service allows users to generate their own customized probe sets

with options for the number of mismatches allowed during

Table 2. The number of raw complete coding sequences extracted for

each organism when considering RefSeq only, or RefSeq and GenBank

Organism Number of Records

RefSeq RefSeq &

GenBank

Arabidopsis thaliana 30 443 88 580

Bos Taurus 7015 10 018

Caenorhabditis elegans 23 122 24 315

Canis familiaris 781 995

Danio rerio 10 957 15 114

Drosophila melanogaster 19 854 22 468

Gallus gallus 4069 4805

Glycine max 0 677

Homo sapiens 24 413 73 885

Hordeum vulgare 0 144

Macaca mulatta 371 656

Medicago truncatula 0 177

Mus musculus 20 286 47 883

Oryza sativa 26 768 28 778

Plasmodium falciparum 0 185

Pseudomonas aeruginosa 0 1

Rattus norvegicus 10 171 15 522

Saccharomyces cerevisiae 0 77

Saccharum officinarum 0 176

Staphylococcus aureus 0 2

Sus scrofa 1182 1953

Triticum aestivum 0 879

Vitis Vinifera 0 152

Xenopus laevis 0 12 004

Zea mays 0 953

The coverage for some organisms (e.g., Rat and Rice) is low. That low coverage

could result in (1) probes being discarded because of no target transcript; and

(2) probe sets failing to be split because of the low resolution. We expect those

problems to be less severe as coverage becomes more complete.

Fig. 2. AffyProbeMiner process flow pipeline. Red font indicates novel

features of AffyProbeMiner, which uses Perl and BLAT to circumvent

the computational bottleneck (indicated by R in the bottle) of the

R program package altcdfenvs. The R computation generates CDFs for

only the Bioconductor environment. In contrast, AffyProbeMiner

generates CDFs for (1) the Bioconductor environment, (2) Affymetrix

GCOS and (3) dChip. The background of Figure 2 is a screenshot of the

web interface.

H.Liu et al.

2388

by guest on May 29, 2013

http://bioinformatics.oxfordjournals.org/

Dow

nloaded from

sequence alignment and minimum number of probes per

probe set. AffyProbeMiner is computationally efficient and

also user-friendly.Figure 3 shows the distribution of probes on one human

chip, HG-U133A, that map to multiple records. The results

shown in Figure 3 indicate that neglecting the GenBank

complete coding sequences will lead to an overly optimistic

estimate of the number of transcript-unique probe sets.There are a total of 25 582 redefined probe sets for the

human U133A chip. Of those, 6878 map to single complete

coding sequences, 16 426 map to multiple complete coding

sequences of a single gene, and the rest map to multiple genes.

Figure 4 summarizes the results of the remapping process. The

peak at (1, 11) represents the relatively small percentage

(27.6%) of original probe sets as designated by Affymetrix

that were not affected by the remapping process; the majority

(72.4%) were, in fact, affected. Strikingly, 70.5% of the

redefined probe sets that map to a single gene are not

transcript-unique (i.e. the probes in them map to multiple

splice variants). The most common reason, in fact, that

redefined probe sets end up non-unique is mapping of probes

to multiple splice forms, rather than to different genes.

AffyProbeMiner offers a number of significant advantages

over other packages (Table 3). Included are coverage of all

Affymetrix chips except exon chips (see subsequently),

an annotation package, a web server for custom computation

or remapping of custom chips, remappings based on high

quality complete coding sequences (i.e. excluding ESTs but

including both GenBank and RefSeq entries), an option for the

user to select either ‘consistent’ (less stringent) or ‘unique’

(more stringent) sets, and an option for the user to select

the minimum number of probes required in the remapped

probe sets. Those features, along with the computational

tractability and user-friendly, integrated web interface, make

AffyProbeMiner a comprehensive solution for the remapping

problem.

4 FUTURE DIRECTION AND CONCLUSION

A relatively new technology, the exon chip, is available from

Affymetrix, currently for human, mouse, and rat.

Conceptually, it is similar to the traditional Affymetrix chips

in that there are ‘probe sets’. However, the probe sets are

directed toward coverage of exons, rather than the 30 ends of

genes. AffyProbeMiner does not currently support the new

exon chips, but we are in the process of extending its processing

logic to do so. We expect to integrate exon chip features into

the website in the near future.In separate studies that are beyond the scope of this article,

we are using AffyProbeMiner to study several biologically and

biomedically important research problems, including enhanced

characterization of theNCI-60 cancer cell line (Weinstein, 2006).In conclusion, the AffyProbeMiner web site (http://discover.

nci.nih.gov/AffyProbeMiner)

� Enables the user to download pre-computed, redefined

CDFs that are compatible in format with most

Fig. 3. The distribution of HG-U133A probes that are mapped to

multiple records. Two mapping strategies are considered: (1) RefSeqs

only and (2) RefSeqs and GenBank complete coding sequences.

The relative number of transcript-unique probe sets is given by the

ratio of frequencies evaluated at abscissa¼ 1. Neglecting the GenBank

records will lead to an overly optimistic estimate of the number of

transcript-unique probe sets. Fig. 4. Three-dimensional histogram of the number of probes in

redefined probe sets and number of original probe sets contributing

probes to the redefined probe sets. A few examples will help in

interpreting this figure:

(i)The large peak at (1,11) indicates the number of instances in which

the redefined probe set was identical to an Affymetrix-defined probe

set. That is, the Affymetrix-defined probe set, containing 11 probes,

gave rise to a redefined probe set containing 11 probes (i.e. all 11 probes

in the redefined set arose from one Affymetrix-defined set.)

(ii)The large peak at (1,1) indicates the number of instances in which the

redefined probe set was a singleton, and of necessity arose from a probe

from one Affymetrix-defined probe set.

(iii)Another special case is the point (2,22). All 22 probes in two

Affymetrix-defined probe sets were merged into a single redefined

probe set.

(iv)A somewhat more general case is the point (2,6). The six probes in

the redefined set arose from 2 Affymetrix-defined probe sets: p probes

from one Affymetrix probe set and (6� p) probes from a second

Affymetrix probe set.

AffyProbeMiner: a web resource for Affymetrix

2389

by guest on May 29, 2013

http://bioinformatics.oxfordjournals.org/

Dow

nloaded from

microarray data analysis platforms, includingBioconductor (Gentleman et al., 2004), AffymetrixGCOS, and dCHIP.

� Provides programs with which the user can construct

redefined CDFs for any short oligo custom arrayhaving Affymetrix-format CDFs.

� Deploys a server that can upload probe sequence filesin FASTA format for generating redefined CDFs.

� Includes a number of important algorithmic, computa-

tional, and usability features that differentiateAffyProbeMiner from other available approaches tothe remapping problem.

ACKNOWLEDGEMENTS

This research was supported in part by the Intramural Research

Program of the NIH, National Cancer Institute, Center forCancer Research, by NSF grant IIS-0430743 to HL, and by

the Intramural Research Program of the NIH, Center for

Information Technology and the NIH Clinical Center. Fundingto pay Open Access charges was provided by NSF research

grant IIS-0639062 and NIH Intramural Research Program,NIH/CCR.

Conflict of Interest: none declared.

REFERENCES

Boue,S. et al. (2003) Alternative splicing and evolution. Bioessays, 25, 1031–1034.

Breitling,R. (2006) Biological microarray interpretation: the rules of engagement.

Biochim. Biophys. Acta, 1759 (7), 319–327.

Carter,S.L. et al. (2005) Redefinition of Affymetrix probe sets by sequence

overlap with cDNA microarray probes reduces cross-platform inconsistencies

in cancer-associated gene expression measurements. BMC Bioinformatics,

6, 107, 40.

Dai,M. et al. (2005) Evolving gene/transcript definitions significantly alter the

interpretation of GeneChip data. Nucleic Acids Res., 33 (20), e175.

Elo,L.L. et al. (2005) Integrating probe-level expression changes across

generations of Affymetrix arrays. Nucleic Acids Res., 33 (22), e193.

Gautier,L. et al. (2004) Alternative mapping of probes to genes for Affymetrix

chips. BMC Bioinformatics, 5, 111.

Gentleman,R.C. et al. (2004) Bioconductor: open software development for

computational biology and bioinformatics. Genome Biol., 5 (10), R80.

Harbig,J. et al. (2005) A sequence-based identification of the genes detected by

probesets on theAffymetrix U133 plus 2.0 array.Nucleic Acids Res., 33 (3), e31.

Hornik,K. (2007) Frequently asked questions on R. Available at: http://cran.r-

project.org/doc/FAQ/R-FAQ.ps.gz.

Kent,W.J. (2002) BLAT–the BLAST-like alignment tool. Genome Res.,

12, 656–664.

Kong,S.W. et al. (2005) CrossChip: a system supporting comparative analysis of

different generations of Affymetrix arrays. Bioinformatics, 21, 2116–2117.

Lee,C. and Roy,M. (2004) Analysis of alternative splicing with microarrays:

successes and challenges. Genome Biol., 5 (7), 231.

MAQC Consortium, (2006) The MicroArray Quality Control (MAQC) project

shows inter-and intraplatform reproducibility of gene expression measure-

ments. Nat. Biotechnol., 24, 1151–1161.

Stoughton,R.B. (2005) Applications of DNA microarrays in biology. Ann. Rev.

Biochem., 74, 53–82.

Weinstein,J.N. (2006) Spotlight on molecular profiling: Integromic analysis of the

NCI-60 cancer cell lines. Mol Cancer Ther., 5 (11), 2601–2605.

Table 3. System comparison

Attributes Authors/Tools

Gautier Dai AffyProbeMiner

Remapped CDFs down-loadable No Yes Yes

All Affymetrix chips included No Yes Yes

Annotation Package No No Yes

Web server for custom computation

or custom chips

No No Yes

Software down-loadable Yes No Yes

Computationally efficient (i.e., does

not require multimode cluster)

No NA Yes

Uses only complete coding sequences

(i.e., no ESTs)

No No Yes

Option to keep ’consistent’ sets or

‘unique’ sets*

No No Yes

Option to set threshold for number of

probes per probe set

No No Yes

User-friendly integrated web interface No Yes Yes

NA, not applicable.

*We define a transcript-consistent probe set as one in which each probe maps to

the same set of transcripts, and a transcript-unique probe set as a transcript-

consistent probe set in which each probe maps to a single transcript. This table is

intended for comparing Gautier with AffyProbeMiner and Dai with

AffyProbeMiner, not for comparing Dai with Gautier.

H.Liu et al.

2390

by guest on May 29, 2013

http://bioinformatics.oxfordjournals.org/

Dow

nloaded from